Introduction

Rice is considered as an important crop globally, providing the main food for more than half of the world’s population, which depends on it. There are many parameters that can affect sustainable rice farming, including timely and appropriate pest control to keep the optimum yield. The potential of remote sensing makes it useful as a time saving and low-priced system to recognize unhealthy plants in a ground scale, primarily for the reason that infected plants contain various spectral reactions in comparison to healthy ones (Zhang 2002).

Providing a significant indicator of a plant’s health status, which may depend on the habitat quality of the specific environment, is the level of the chlorophyll content of vegetation. Organisations can improve the monitoring of vegetation properties by means of current developments in optical remote sensing.

The natural pigment in vegetation is chlorophyll (mostly in leaves) which has an important role in evaluating a plant’s biological condition. Chlorophyll content (Chl) acts as an indicator of an essential procedure for growth and survival known as photosynthesis (Peng and Gitelson 2012), and it is as an indicator of transformations, stress and nutritional situation as well (Evans 1989). Therefore, knowing about the amount of chlorophyll per unit leaf area is important in environmental, agricultural or climate researches because of its direct relation to atmospheric carbon fixation, and therefore biomass accumulation (Gianquinto et al. 2011).

To evaluate the potential application of new images to detect ground surface changes and also for weather conditions it was necessary to promote change detection and threshold methods on the images. Morphological plant issues indicating damage were accessible at the canopy due to inner destruction of chlorophyll pigments and tissue construction for photosynthesis and metabolism while plants were infested with pests. Therefore, the diseased plants will have different individual spectral structures from healthy plants. For discriminating this spectral contrast and recognising the diseased plants or patches in the field, remote sensing can be used (Zhang et al. 2003).

The main aim of this study is to evaluate the effectiveness of integrated pest management on environmental and weather foundations for generating a proposal and awareness for agriculturalists about rice pests and furthermore, produce an early warning system for BPH detection in rice farming by means of remote sensing to assess chlorophyll content and weather data.

Symptoms of BPH will not be visible from outside in the early stages, but if we enter the field and tap the plants many times, this insect can be seen. They are visible only when the damage is severe; indeed, the plants present a burned up appearance—so called hopper burn—in circular patches. Symptoms of BPH appear more in the later stages of plant growth, when they turn from a green colour to brownish very quickly.

Studies mention that the BPH nymph can survive at the greater limits of tolerable temperatures. This suggests that climate warming in tropical districts with high temperatures (35–45 °C) would support the survival and spreading of BPH. Some weather parameters such as temperature, RH and Ra act as main boosters in the growing of the pest population. To assess the possible presentation of new images to distinguish ground surface variations and also for weather condition require endorsing threshold procedures and change detection on the images. Remote sensing (RS) is an advanced implement to detect land surface processes on a large scale and in a cost-effective approach (Schmugge et al. 2002).

Satellite multispectral sensors that monitor the greenness of vegetation are generated by means of the visible and near infrared (NIR) bands. Reflection of the NIR band in stressed vegetation, which in visible band has no energy absorption, is less than in non-stressed ones. Consequently, for observing the influence of drought on vegetation in moisture stressed and standard crops, wavelengths are more appropriate.

BPH exists for about 25–28 days in temperatures of 27–33 °C, and the cycle consists of three growth stages: egg, nymph and adult. The ecology of BPH life will be changed in relation to weather conditions. The most important factor that influences the transfer from one growth stage to the other is temperature (Pathak 1968; Ho and Liu 1969; Bae and Pathak 1970; Kalode 1976). If the temperature is about 25–30 °C and humidity is from 80 to 86 %, BPH density (individuals/m2) rates after changing to new growth stage increase between 50 and 60 % (Nguyen et al. 2006).

Consequently, this research addresses the issue of the extraction of the conducive environment for BPH i.e. temperature, NDVI, relative humidity (RH) and extra-terrestrial radiation (Ra) using RS data to establish an early warning technique for pests in rice farming.

Methodology

Study area

Tanjung Karang Rice Irrigation Scheme is positioned at about 3°25′ to 3°45′N and 100°58′ to 101°15′E in the state of Selangor Malaysia. Rice is grown twice a year mostly from August to January (main or wet season) and February to July (off or dry season).

Sawah Sempadan compartment consists of 1468 lots with the total area at about 2300 hectares, divided into 24 blocks. Block C in Sawah Sempadan compartment had 86 individual farmers and was chosen as the main study area. The research area ‘Block C’ located at 3°28′09.63465″N 101°13′26.48399”E with an average altitude of 6.2.0 m above mean sea level (Fig. 1).

Fig. 1
figure 1

Sawah Sempadan Compartment and Block C. a Satellite image for Tanjong Karang Rice Irrigation Scheme, b Satellite image for Sawah Sempadan Rice Irrigation Compartment, and c Block C at Sawah Sempadan rice irrigation compartment

Full size image
Fig. 2
figure 2

SPOT-5 subset image for Block C

Full size image
Fig. 3
figure 3

BPH trapped by sticky paper

Full size image

Image data information

SPOT-5 (Satellite Pour l’Observation de la Terre) images were collected within 101°12′42.82″N and 3°28′49.88″E at upper left and also 101°13′49.71″N and 3°27′35.57″E at the lower right side (Fig. 2). The images had four bands, in green: band 1 (530–600 nm); red: band 2 (610–680 nm), NIR: band 3 (780–1000 nm) and mid infrared: band 4 (1000–2500 nm). Spatial resolution of the images was 2.5 m. Images provided for five different years consisted of: August 2007, August 2008, January 2010, 9 March 2012 and 10 March 2012. Most of the time, at the late growth stage BPH infection is greater. By this means, these dates of imagery were selected which are in both growth stages for comparison. The images were kept as 8-bit digital number (DN) values ranging from 0 to 255. Landsat-8 images for four different dates: March 2012, January 2010, August 2008 and August 2007 were collected for extracting the weather parameters. The bands used were thermal bands 6.1 and 6.2, with 30 metres resolution.

Image processing

Image data were extracted using ENVI 4.8 for producing subset captures of the field from the Landsat and SPOT-5. Initially, BPH sheath blight through the paddy field was detected to identify unhealthy areas by SPOT-5 images in previous research (Faranak et al. 2014); the ENVI spectral utility was applied for data extraction from the subset images. Converting DN to radiance was completed and Landsat gain and offset were determined. Among the processes, masking the clouds through subset images was performed. By this means, analyses were undertaken for discriminating the relationships among plant greenness which shows by NDVI, temperature and precipitation which is extracted by Landsat-8. The data set has satisfied several quality criteria and has been used in several analyses for extracting temperature, NDVI, RH and Ra from SPOT-5 and Landsat-8/ETM+ Thermal-Band images. Methodology was constructed on a fixed NDVI threshold value to identify healthy plants from unhealthy plants (Ribbes and Le Toan 1999).

Temperature, rainfall and humidity data were collected from a nearby station as the reference. By use of relevant formula (Abduwasit 2010) (Table 1) in the previous research, extracting the weather parameters were completed with Landsat ETM thermal bands. K1 = C1/ʎ5, and K2 = C2/ʎ, and satellite measured radiant intensity Bʎ (T) = Lʎ. Where C = 1.19104356 × 10−16 Wm2; C2 = 1.43876869 × 10−2 m K and ʎ is wavelength.

Table 1 Amount of K for Landsat
Full size table

K1 and K2 become a coefficient specified by effective wavelength of a satellite sensor consist of NIR and MIR bands.

Probability analysis

By use of ArcGIS, maps for Temp, RH and Ra by the kriging method have been created for discriminating healthy plants from unhealthy ones. On the other hand, for specifying infected parts from non-infected ones, BPH maps for each date were created as the base. The data was prepared by checking the pixel and boundary for example, and making them into a layered map for transferring to an ASCII database; raster datasets were converted to ASCII files using a raster calculator. All the maps were prepared so they could be used in SPSS analysis. Using read text data, the text was opened; specific numbers of variables were allocated number 1 and names were defined for each of them. Missing values were designated as −9999 and defining the measure as a scale, they were merged. They were prepared for running in a binary logistic regression model (LR), while the training data and the parameters were discriminated for creating suitable equations. SPSS software was used to perform the multivariate LR analysis. The higher the logistic coefficient, the more impacts on occurrence of rice will be expected (Ayalew and Yamagishi 2005). The probability map can be generated using the derived logistic coefficient through the formula that can be expressed as:

$$ P = 1 / (1 + e^{ - z} ) $$
(1)

where p is the probability of flooding which is acquired between 0 and 1 on an S-shaped curve. Z is linear combination and it follows that LR involves fitting an equation of the following form to the data:

$$ Z = b^{0} + b_{1} x_{1} + b_{2} x_{2} + b_{3} x_{3} + b_{n} x_{n} $$
(2)

where b0 is the intercept of the model, bi(i = 0,1,2,…, n) denotes the LR coefficients, and xi(i = 0, 1, 2,…, n) indicates the conditioning factors (Lee and Sambath 2006). Eventually, a proper decision was made by R-square achievement in a separated form of all the data.

Data verification

Data collection from block C in Tanjung Karang area were conducted during the first planting season of the year (January to April). Yellow papers and sticky glue to trap the BPH were punched on wooden sticks and located uniformly throughout the study area (Fig. 3). They were collected and replaced with a set of yellow papers and the amount of BPH was established. The coordinates of the sampling locations were detected by using GPS.

BPH maps were generated in ArcGIS by Kriging and raster calculator methods under spatial analysis tools, then, illustrating the BPH sensitivity graph specifying the estimated attacked regions.

Model validation

The area under the curve (AUC) is very popular method due to its comprehensive, understandable and visually attractive way of accuracy assessment. Prediction and success rate should be evaluated as an essential outcome of every program. The validation process was implemented by comparing the existing data with the acquired BPH probability map. The model was validated using quantitative method of AUC in which the basis of the assessment is the true- and false-positive rates (Tien Bui et al. 2013). The range of the AUC varies from 0.5 to 1.0, as the value of 1.0 represents the highest accuracy showing that the model was completely satisfied to predict the disaster occurrence without any biased effect. So the closer the value of AUC to the 1.0, the model is considered to be more precise and trustable.

Results

Regarding the previous investigation, results showed that there were dissimilarities among the images between early and late growing seasons, looking at particular image indices; specifically RVI 14, SDI 14 and SDI 24, which indicated better correlations for detecting healthy plants from diseased ones. These types of indices could be valuable indicators for an emerging technique to detect the sheath blight of rice via remote sensing (Faranak et al. 2014).

The results of correlation between the Landsat image and station data for average temperature and RH for four different months indicate August 2008 and March 2012 show higher temperature magnitudes in comparison with two other months, although the RH average is slightly higher in January 2010 and March 2012 compared with the rest.

By using LR model, the correlations for the data from the stations and Landsat-8/ETM were estimated; the correlation and regression is shown on the following Table.

Correlation results for temperature and RH data which were extracted by Landsat images have shown in Table 2 that all of them are significant at 0.01. Based on the regression results (R2 value) on Table 2, extracted temperature from the Landsat image had a significant positive association with the station values.

Table 2 Correlation and regression analysis of Landsat for 120 parameters
Full size table

SPSS analyses and ArcGIS maps

By comparing the thermal band image data i.e. NIR and MIR with any vegetation index, such as NDVI or the greenness of the plant, the dependence of surface temperature on the scale of biomass could be estimated statistically (Maskova et al. 2008). The Landsat image bands were turn into an ASCII dataset, then imported for statistical analysis.

The resident intensity of radiation is, nevertheless, specified by the angle along the direction of the sun’s rays and the angle to the surface of the atmosphere. This defined angle will modify through the day and will be different at various latitudes and in different seasons. The extra-terrestrial radiation, Ra, for different latitudes and for every day of the year can be computed from the solar constant, the time of the year and the solar declination. Actually, all the latitude values used in the formulae for obtaining Ra were similar, so the maps which were created for Ra were not useful for discriminating the parts because the coordinates were very similar and the differences were few. Consequently, Ra maps were not useful for this aim.

Temperature is likely the most important environmental issue influencing insect’s behaviour, distribution, growth, existence and reproduction because they are ectothermic organisms, the temperature of their body converts almost with the temperature of their habitats (Chu and Chao 2000). Higher temperatures in some insect pests are likely to motivate the breeding of adults and lead to quicker population growth (Harrington et al. 2001). Warmer situations even in regions with climate that is already warm is generally a better habitat for insects (Frazier et al. 2006).

Warm and humid environment is more pleasing for several pest species. Vulnerability of crops to pest damage is mostly due to both direct and indirect effects of moisture stress on them. There are not many scientific surveys on how precipitation would affect insects. Nevertheless, heavy rains can kill or remove some insects from crops which is caused by their sensitivity to precipitation. A winter rainfall reduction could result in declined aphid developmental rates (Pons et al. 1993); reduction of reproductive capacity of overwintering aphids was due to drought-stressed tillage cereals (Pons and Tatchell 1995).

Model development

Data was prepared for running the model while the training data and the parameters were discriminated for, to make suitable equations. Eventually, a proper decision was made by R-square achievement in a separated form of all the data.

Research on pests during long-term monitoring of the rice insect pests in Taiwan suggests that inhabitants of various insect pests, consisting of the immigrated and native species, have increased significantly over the past 10 years (Huang et al. 2009a). Global warming is a highly significant issue and will possibly play a role in the increasing of their populations. Detecting the population of these insect pests is very important for pest management, to consider a probable change of rice pests in the next years (Huang et al. 2009b). In order to produce probability map the values obtained from LR were transferred to the ArcGIS software and Eq (2) was applied to the relevant factors.

Resulted coefficient for the data of “Aug 2007”:

$$ ( 4. 9 7 3 8 9 1 \times {\text{``(NDVI}} ) {\text{2007''}} ) - ( 0. 3 1 8 5 3 5 \times{\text{``rh2007'') + (0}} . 0 7 4 0 3 0 \times {\text{``temp2007'') + 6}} . 6 8 4 0 0 0 $$
(3)

Resulted coefficient for the data of “Aug 2008”:

$$ ( 0. 2 3 2 6 2 9 \times {\text{``temp2008'')}} - ( 1 1. 5 0 4 0 5 1 \times \left( {\text{``NDVI}} \right) {\text{2008''}} ) - ( 0. 0 7 6 8 3 4 \times {\text{``rh2008'')}} - 3 6. 9 9 7 4 0 7 $$
(4)

Resulted coefficient for the data of “Jan 2010”:

$$ ( 3. 7 2 3 6 1 9 \times {\text{``(NDVI)2010''}} )- ( 0. 1 0 2 3 1 0 \times {\text{``rh2010'')}} - ( 0. 2 0 9 7 2 6 \times {\text{``temp2010'') + 50}} . 0 9 6 7 3 3 $$
(5)

Resulted coefficient for the data of “Mar 2012”:

$$ ( 6. 6 5 9 0 2 6 \times {\text{``(NDVI}}) 9 {\text{th'')}} - ( 0. 1 6 4 8 3 9 \times {\text{``rh9th'') + (0}} . 5 0 2 8 1 0 \times {\text{``tem9th'')}} - 9 7. 3 3 1 1 7 5 $$
(6)

Equation (5) 10 Mar 2012:

$$ ( 0. 2 3 2 1 5 1 \times {\text{``temp10th'') }} - ( 0. 0 0 8 5 7 4 \times ( {\text{``NDVI''}}) 1 0 {\text{th'') }} - ( 0. 0 6 9 8 5 4 \times {\text{``rh10th'')}} - 4 3. 8 4 1 5 3 1 $$
(7)

Maps for each date and parameters

With the aim of achieving an accurate result, relevant maps comparisons, i.e. NDVI, T, RH and BPH layouts, have been conducted for five different years and the results shown below indicate the probable regions more heavily attacked by BPH. Figures illustrate the BPH sensitivity graph specifying the estimated attacked regions for different years.

Figures 4 and 5 illustrate the BPH sensitivity graph specifying the estimated attacked regions within two selective years i.e. 2010 and 2012. Furthermore, BPH probability maps, as shown in Figs. 6, 7, 8, 9, and 10, were used to determine the probability of BPH infestation locations, with which the visual similarity with field collection maps is obvious. Moreover, Kappa coefficient was applied and it showed 0.80 (± 0.05) for all the years which represent an acceptable correlation between BPH infestation locations and the ‎‏estimated one‎.

Fig. 4
figure 4

Map of BPH attacked regions in the field for January 2010

Full size image
Fig. 5
figure 5

Map of BPH attacked regions in the field for March 2012

Full size image
Fig. 6
figure 6

Map of BPH probability regions in August 2007

Full size image
Fig. 7
figure 7

Map of BPH probability regions in August 2008

Full size image
Fig. 8
figure 8

Map of BPH probability regions in January 2010

Full size image
Fig. 9
figure 9

Map of BPH probability regions on 9 March 2012

Full size image
Fig. 10
figure 10

Map of BPH probability regions on 10 March 2012

Full size image

By describing Fig. 6, can conclude that BPH distribution is wide in the field, and also, it is more obvious in upper and lower parts.

Figure 8, is the map which shows comparisons, i.e. NDVI, T, RH and BPH layouts, and indicates that BPH spreads more in the upper and middle parts of the field.

Here in Fig. 10, can see that the greatest BPH attack occurred in the middle part of map and also most population are in upper-west corner rather than the other parts, which is almost similar to Fig. 5.

Usually, for finding meteorological data for energy stability models, thermal remote sensing has been utilized (Snyder et al. 1998). This type of technique was used for extracting the particular time-of-day components of the energy balance at the time of the remotely sensed data acquisition. The fact that climate data collected from stations, including maximum and minimum temperatures and daily precipitation, can be exactly connected with thermal remote sensing such as multispectral or AVHRR (Advanced Very High Resolution Radiometer) data was found by McVicar and Jupp (1999). Consequently, these data, by permitting better definition of the spatial energy balance, can be utilized to increase the spatial density of meteorological stations. An extremely powerful correlation between the minimum air temperatures and the thermal band satellite-based surface temperature was found by comparing them with each other (Wang and Li 2012). These kinds of data can also be connected with land cover data, therefore a complete biophysical and meteorological analysis of a landscape is supplied (François et al. 1999).

With the knowledge that the late growth stage of plants has more severe BPH attacks, by visual and via results of correlation and regression values, the results indicated that BPH outbreak is particularly obvious in the north-west corner and middle regions of the maps and it is more likely to happen in specified range of temperature and RH, i.e. 29 °C <T< 32 °C and 88 % < RH < 93 %.

Model for an early warning system

The training data and the parameters have been identified for constructing suitable equations to run the model in the same way as the other analysis. Finally, an appropriate model was made by R-square achievement for BPH: The prediction rate shows how well the model can predicts the BPH in an area. For that reason, 30 % (4 points from 13) of attacked locations that was saved for the purpose of testing were used to measure the prediction rate. So the model’s generalization ability can be measured through this method. The LR model produced the value of 0.845 for area under the curve (AUC) which represents 84.5 % success accuracy. The prediction curve is illustrated the AUC of 0.815, which corresponds to the prediction accuracy of 81.5 %.

Several parameters of weather conditions like humidity (moisture), sunlight hours, temperature, and heavy rainfall influence the main effects on the concentrations of pest population. Rainfall and temperature are the most significant and simple to retrieve among the weather factors that can be remotely sensed. To assess the possible presentation of new images to distinguish ground surface variations and also for weather conditions requires endorsing threshold procedures and change detection on the images. BPH can exist for about 25–28 days in temperatures of 27–34 °C and consist of three growth stages: egg, nymph and adult. In relation to the weather conditions, the ecology of BPH will be changed.

Results of model verification

Based on this model and analysis it can be concluded that BPH can be alive at the higher limits of acceptable temperatures and RH. Therefore, it states that in March while the plants are between 70 and 90 days of the growth stage, high temperature up to 32 °C and also high RH up to 93 % causes increases in distribution and survival of BPH. Transferring from one growth stage to the other for BPH was caused by temperature. Based on the models temperature and RH are two significant factors which can affect BPH infestation in the field, particularly in March when the weather becomes warm.

Conclusion

In order to detect BPH in the field and using remote sensing data in pest detection, satellite images with thermal bands suggest precise, valuable, unique and sustainable resources of detecting environmental factors and changes and they can also assess the spatial features of the plants and vegetation’s energy budget (Southworth 2004).

Usually, for finding meteorological data for energy stability models thermal remote sensing has been utilized (Snyder et al. 1998). For extracting the particular time-of-day components of the energy balance at the time of the remotely sensed data acquisition this technique was used. The fact that climate data which is collected from stations, and includes maximum and minimum temperatures and daily precipitation, can be exactly connected with thermal remote sensing such as Landsat or advanced very high resolution radiometer (AVHRR) data was found by McVicar and Jupp (1999). Consequently, these data can be utilized by permitting better definition of the spatial energy balance to increase the spatial density at meteorological stations. An extremely powerful correlation between the minimum air temperatures and the thermal band satellite-based surface temperature was found by comparing them with each other. These kinds of data can also be connected with land cover data, therefore a complete biophysical and meteorological analysis of a landscape is supplied (François et al. 1999).

The study showed that SPOT images could be used to prove the coarse scattering of infected plants in the field by recognizing chlorophyll content and the degree of green colour in the leaves which is caused by temperature. Examination shows that nymphs of BPH are existing at the higher limits of sustainable temperatures and RH. As a result, in tropical areas commonly in March when the climate becomes warm with infrequent, particularly high temperatures and high RH, BPH infection would be more than at other growing stages. Transferring from one growth stage to the other for BPH was caused by temperature.

From the results, one can conclude that in the March when temperature and RH become high, BPH infestation would be increased as well. Consequently, it can be a good reminder for farmers to prevent BPH attacking before March with common methods, i.e. by using some pesticide.

This analytical study should be developed into more cases to obtain the best agreement for detecting pests and diseases in agricultural fields by use of different remote sensing devices. In addition, by having more output data, an applicable parametrical equation can be found with the aim of predicting weather factors which can more effective on pests of rice.